import torch
import torch.nn as nn
import random
from model.tokenizer import Tokenizer

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
tokenizer = Tokenizer()
VOCAB_SIZE = 10000  # Temporary cap, grows dynamically
EMBED_DIM = 128


class TinyTransformer(nn.Module):
    def __init__(self):
        super().__init__()
        self.embed = nn.Embedding(VOCAB_SIZE, EMBED_DIM)
        self.ln1 = nn.LayerNorm(EMBED_DIM)
        self.fc = nn.Linear(EMBED_DIM, VOCAB_SIZE)

    def forward(self, x):
        x = self.embed(x)
        x = self.ln1(x)
        return self.fc(x)


model = TinyTransformer().to(DEVICE)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
loss_fn = nn.CrossEntropyLoss()


def generate_response():
    seed = torch.tensor([random.randint(0, tokenizer.next_id - 1)], device=DEVICE)
    output = model(seed.unsqueeze(0))
    pred = torch.argmax(output, dim=-1).squeeze().tolist()
    if not isinstance(pred, list):
        pred = [pred]
    return tokenizer.detokenize(pred)